Why Singapore schools need AI policies now: A chemistry teacher’s warning

We’re letting students use AI without clear boundaries. Here’s what I’m seeing in my classroom, and why administrators need to act before this becomes a crisis. 

Last year, a student showed me an artificial intelligence (AI)-generated answer to a 2024 A-Level Chemistry question asking why calcium fluoride remains insoluble despite a negative Gibbs free energy solution. 

ChatGPT delivered a long, technically accurate explanation about kinetics and lattice energy. It sounded impressive. It was also wrong for the exam.

The expected answer was simply that the activation energy is too high. The student had everything except the one idea that mattered. 
The AI answered it, they copied it, and they never questioned whether it made sense. This isn’t a one-off situation. If we don’t set clear rules around this soon, we’ll stop measuring what students actually know. 

We’re operating without a playbook 
Right now, most schools are winging it. 

Some teachers ban AI completely. Others let students use it freely. Many cross their fingers and hope kids will figure out “responsible use” on their own, whatever that means. This isn’t working, and it’s creating real problems. 

When students hand in work, and we can’t tell how much of it is theirs, we’ve lost our ability to spot who’s struggling and who needs help. Academic integrity becomes impossible to enforce when nobody’s clear on what counts as cheating versus legitimate help. And weirdly, students in the same school are getting wildly different educations depending on whether their teacher is pro-AI or anti-AI. 

More importantly, students lose the struggle that makes learning stick. Wrestling with why thermodynamics and kinetics don’t always align is the mental work that builds understanding. When AI delivers an answer fully formed, that process disappears, and with it, authentic learning. 

What good AI policies actually look like 
From what I’m seeing on the ground, schools need to split AI use into three clear buckets: things you absolutely can’t use it for, things you can use it for with oversight, and things where it actually helps. 

For the no-go zone, I’d include any assessed work and homework intended to build core skills. In chemistry, that means drawing mechanisms, doing calculations, explaining concepts, the fundamental stuff. These aren’t pointless exercises. They’re how students develop the intuition they’ll need later when AI gives them garbage, and they need to spot it.

Students who let AI do all their mechanics practice never build that gut feeling for what looks right. And you can’t just download that intuition later. You have to earn it through repetition and mistakes. 

The supervised category could include things like reviewing concepts after you’ve learned them, double-checking your work, or generating extra practice questions. However, and this is critical, “supervised” has to mean something. 

Teachers need to actually check whether students understand the AI-generated material or if they’re copying it mindlessly. That means we need to train teachers to recognise AI-generated content and probe for a fundamental understanding, which most of us honestly aren’t equipped to do yet. 

The encouraged uses should be tasks where AI genuinely adds value: looking up scientific papers, visualising complex molecules, and exploring hypothetical scenarios. The catch is timing. These only work after students have built enough foundational knowledge to evaluate the AI’s output. 

How to actually make the policy work 
Writing a policy is the easy part. Getting it to work in practice? That’s where things get messy, and administrators need to plan for some predictable headaches. 

First, forget about relying on AI detectors. They’re unreliable, they flag innocent work and miss obvious AI content. Plus, students adapt fast. They learn to tweak AI outputs until they pass detection. 

We can’t technology our way out of this. Instead, we need to redesign how we assess, such as by doing more in-class work, more oral exams, and more questions that ask students to explain their reasoning step by step. 

Second, teachers are all over the map on this. Some see AI as the end of education as we know it. Others think students need to use it constantly to prepare for the real world. Professional development needs to help everyone understand both the real limitations of AI and its legitimate uses. 

Teachers need practical training in creating assignments that AI can’t easily complete and in using AI as a teaching tool when it makes sense. 

Third, parents complicate things. Many bought ChatGPT subscriptions specifically for homework help. When we tell them their kids can’t use it, we need a better explanation than “because we said so.” 

We have to help parents understand that real understanding comes from effort, not from having answers handed to you. It’s the difference between watching someone do pushups and doing them yourself. 

They’ll use it at work anyway 
I hear this argument a lot from administrators that students will use AI in their careers, so we should let them use it freely now. But this gets it backwards.

Using AI effectively at work requires knowing when to trust it and when to question it. It requires enough expertise to catch its mistakes. It requires the ability to function when the AI breaks or encounters something new. You can’t develop any of that if you never learned the fundamentals in the first place. 

Take chemistry careers. Yes, routine calculations and data processing will be automated. But what companies actually need are chemists who can troubleshoot weird lab results, make judgment calls on safety issues, and solve problems that don’t have obvious answers. Those skills develop through the kind of deep thinking that AI shortcuts prevent. 

Future chemistry graduates will likely be hybrid professionals, comfortable with both lab work and computational tools. But that only works if they first learn to think like chemists. If you spend your whole education letting AI explain mechanisms for you, you never develop the intuition that comes from drawing hundreds of wrong arrows and learning why they’re wrong. That’s not something you can catch up on later. 

The choice we’re making right now 
Banning AI from schools entirely is pointless and counterproductive. However, we also cannot continue like this, hoping students will somehow figure out the appropriate use through trial and error whilst their learning suffers. 

The question is how to regulate AI in the classroom intentionally and strategically, or whether we let informal policies emerge randomly through inconsistent enforcement and student experimentation. 

Given what’s at stake, that shouldn’t be a hard choice.